Build Your First Bot

— This article helps you understand the recommended path for building a usable bot from scratch: prepare knowledge first, then build the bot, and finally verify and publish.

1. What Quick Build Does

Before you build your first bot, we recommend understanding the suggested onboarding order:

• Use case: You've just onboarded the product and want to quickly get a bot running that can serve customers.
• Purpose: Complete "prepare knowledge → build the bot → verify and publish" via the shortest path, avoiding the feeling of not knowing where to start.

2. Recommended Build Path

1. Step 1: Upload knowledge to the Knowledge Base first

A bot's answers rely on knowledge, so we recommend preparing knowledge in the Library first:

1. Go to "Library › Knowledge › Knowledge Base" and create a knowledge base.
2. Upload business knowledge using QA questions, articles, files, web pages, tables, and more.
3. See Knowledge Base for details.

2. Step 2: Build the bot (choose either method)

• Build in AI Studio: Best for wanting AI to help you build and lower the barrier. Go to "AI Studio › Run AI Studio," click "Build a Bot" or describe your needs; after AI parses them, it generates the configuration as recommendation cards for you to confirm item by item.
• Build manually in Agent Management: Best for wanting to configure everything yourself. Go to "Agent Management › Agent Setting," click "Create Agent," and manually fill in information, prompts, bind knowledge, and so on.

Both methods build the same bot and can be mixed: use the AI Studio to build the skeleton first, then go to Agent Setting for fine-tuning.

3. Step 3: Verify, then publish

1. Run a few test chats in the AI Studio, or click "Debug" on the Agent Setting page to verify.
2. Once confirmed, click "Publish"; the system automatically runs a configuration check.
3. After publishing succeeds, you can deploy the bot to reception channels to serve customers (see the next section).

3. Deploy

After a bot is published, it does not automatically start serving customers—you must also select it in the reception settings of the corresponding channel. A bot can be deployed to three types of channels—Online Service, Call Center, and Ticket Center—and you only need to configure the channels you have actually enabled.

3.1 Online Service

1. In "Online Service › Settings › Reception Plan Settings › Bot Reception," select your published bot as the reception bot.

2. In "Online Service › Settings › Reception Plan Settings › Transfer-to-Human Rules," set the trigger rules for the bot to transfer to a human agent.

3. In "Online Service › Settings › Conversation Settings › Customer Assignment Strategy," set the agents for routing: when a customer initiates an inquiry and meets the preset trigger conditions, the customer is assigned to a designated agent or skill group according to the assignment rules.

4. In "Online Service › Service Apps › FAQ › FAQ Plan," configure personalized FAQs: provide tailored FAQ guidance based on the customer's identity and their stage in the conversation.

3.2 Call Center

1. In "Call Center › Settings › Inbound Routing," configure routing for customers; when a trunk number is specified, assign the customer to an IVR flow.

2. At the transfer-to-bot node in "Call Center › IVR Flow," select your published bot.

3. After the bot conversation ends, fill in the bot variable name in the bot's outbound parameters to assign the bot variable to an IVR variable.

3.3 Ticket Center

1. In "Ticket Center › Ticket Intelligence › Q&A Bot Settings › Q&A Bot Rules," configure the tickets that can be handled by the bot.

2. In "Ticket Center › Business Process › Ticket Routing Settings," set the skill group or agent that handles the ticket after the bot transfers to a human agent.

4. What You Can Do Next

• Want AI to help you build → AI Studio.
• Want to configure it yourself → Agent Setting.
• Answers are wrong and you want to fix them → AI Studio.

Tip: Changes made through building and tuning take effect only after they are confirmed and published; you can test chat freely before publishing.